Vflash MiniMax H3 derivatives

Publication status: no model weights

This repository currently contains no model weights, adapters, quantized tensors, or other MiniMax H3 model payloads. It is reserved for a future Vflash model release after the required distribution authorization is documented.

MiniMax H3 is governed by the MiniMax H3 Community License Agreement, not the Vflash source-code license. The current agreement defines an Applicable Territory that excludes the European Union, United Kingdom, Republic of Korea, and United States, and section V.4 restricts use, distribution, and display of the model and its Outputs outside that territory. A globally accessible derivative repository therefore needs additional authorization or an authoritative upstream license clarification.

Reproducible code and evidence

• Source and Docker runtime: https://github.com/Hansimov/vflash
• Documentation: https://hansimov.github.io/vflash/
• Anonymized benchmarks: https://hansimov.github.io/vflash/generated/benchmarks/
• Model and media license boundary: https://hansimov.github.io/vflash/model-publishing/

The public code pins LightX2V, carries auditable H3 patches, supports single-GPU and TP2 launch contracts, and publishes both successful and rejected optimization measurements. It does not silently download or bundle H3 weights.

Requirements for a future payload

A future release must include the exact base-model revision, conversion command, tensor manifest and integrity digests, modified-file notices, runtime requirements, measured speed and quality boundaries, the full upstream license, and the required NOTICE. It will not be uploaded until its intended territorial and redistribution scope is covered.

Required upstream notice:

MiniMax H3 is licensed under the MiniMax H3 Community License Agreement, Copyright © 2026 MiniMax. All Rights Reserved.